Fábio André Malko
Tax law, white-collar crime, corporate law and research method
About
Attorney, accountant and university lecturer. Doctoral candidate in Law at IDP, Brazil, since 2026. This site gathers what I write and what I read in tax law, white-collar crime and corporate law, with particular attention to the use of artificial intelligence in legal research.
The Portuguese version carries the full contents, including all articles, notes and reading files. This page summarises the research line for readers outside Brazil.
The research problem
Methodological literature prescribes reproducibility for empirical research assisted by large language models: open-weight models, self-hosting, pinned versions, declared parameters, and verification of the same prompt across several models. The empirical literature that classifies legal text at scale meets little of that prescription, for reasons of feasibility, so that the two bodies of work seldom intersect.
The claim
I argue that reproducibility is a measurable property of each variable of the instrument, to be measured and reported field by field.
I call this functional replicability by field. A single study may be fully replicable in some variables while failing in others, so that a blanket statement about a study being reproducible says little about what can actually be reused.
The supporting evidence is my own. On the same corpus, with the same model, on the same day, agreement between automated classification and human adjudication ranged from 0.97 in one field to 0.17 in another, and only four of ten fields met the adopted acceptance criterion. I publish the lowest figure alongside the highest, because the value of the measure lies in showing where the instrument fails.
The gap
A systematic review of the literature, recording a verbatim excerpt and source page for every mention, found no study that submits the same legal protocol, in Portuguese, to a panel of both open and proprietary models with formal measurement of stability. Each of the nearest works falls short of one of those three conditions.
Contents
The research line in progress. A reading index in which each work receives an analytical note of my own authorship. A research curriculum vitae with persistent identifiers and citation format.
Replication material, DOI-registered datasets and interactive panels follow in later phases, as each clears the review that open publication requires.
The Portuguese version of this site is the version of record. Ir para a versão em português.